README.md

Naive Bayes Email Classifier

Historical machine learning project (2019) — a desktop spam/ham email classifier using bag-of-words features and multinomial Naive Bayes.

This is an archived-in-practice learning project, not an actively maintained spam-filter product. The original code relies on older Python and pandas conventions. The repository is kept as evidence of early ML experimentation, not as a production-ready or benchmarked classifier.

What the code contains

• Dataset loading: reads message bodies from local hams/ and spams/ directories. Separate hams_t/ and spams_t/ directories are used for test examples.
• Feature engineering: sklearn.feature_extraction.text.CountVectorizer turns email text into a bag-of-words feature matrix.
• Classification: sklearn.naive_bayes.MultinomialNB trains on the labelled spam/ham examples.
• Evaluation: the original script computes true positives, true negatives, false positives and false negatives, displays a confusion matrix and simple classification statistics.
• User interface: app_final.py provides a Tkinter-based desktop form.

spam / ham messages
         │
         ▼
   CountVectorizer
         │
         ▼
    MultinomialNB
         │
         ▼
    spam / ham label

Important reproducibility limitations

• Some code uses Python 2 syntax, including Tkinter and iterator conventions.
• Deprecated pandas methods such as DataFrame.append are present.
• Some GUI/evaluation routines need fixes before a current Python 3 run.
• There are no independently validated precision/recall numbers claimed in this README.

For the current applied-ML direction, see the Production ML Pipeline roadmap covering leakage checks, measured evaluation, calibration, serving and drift.

Author's portfolio: Omar Laraqui · Selected work